A new staff analytical paper from the Bank of Canada offers one of the more thorough examinations of decentralized lending to date, and its conclusion is both a compliment and a quiet critique.
Aave V3, the largest DeFi lending protocol by total value locked, managed to avoid unrecovered bad debt across the study period. But the way it did so, the paper suggests, involved pushing a significant portion of the risk onto the borrowers it serves.
The paper, titled “DeFi Lending: Returns, Leverage, and Liquidation Risk” and authored by Bank of Canada researchers Jonathan Chiu and Furkan Danisman, studied decentralized lending on Aave V3 using transaction-level data.
It analyzed the protocol’s revenue model, borrower behavior, and liquidation dynamics, finding that while DeFi lending with proper governance is operationally viable, it also faces constraints related to capital efficiency, liquidation risk, and systemic fragility within the crypto ecosystem.
Using transaction-level data from Jan. 27, 2023, to May 6, 2025, the study found that Aave V3 reported zero non-performing loans in 2024, with overcollateralization and automated liquidations helping prevent lender losses in its Ethereum lending market. Positions were typically liquidated before collateral values fell below outstanding debt, helping contain lender losses across the sample.
That’s a meaningful finding for a protocol managing billions in open positions through multiple volatile market cycles. But the paper’s more pointed observation is what comes after that headline number.
The hidden cost: who actually bears the risk
While the model protected lenders from unrecovered losses, it also shifted risk onto borrowers and constrained capital efficiency compared with traditional lending systems. Aave V3’s design relies on automated risk controls rather than traditional underwriting, requiring borrowers to post more collateral than they borrow and liquidating positions when they breach risk thresholds.
This is the fundamental tension at the core of DeFi lending as it currently exists. There’s no credit check, no relationship-based lending, no loan officer using judgment. What there is instead is a set of smart contract rules that kick in automatically, and when those rules are triggered, borrowers can find themselves on the losing end of a rapid, automated position closure before they’ve even had a chance to respond.
The paper estimated that liquidation fees typically ranged from 5% to 10% of liquidated value, while missed gains from subsequent price recoveries pushed combined losses to about 10% to 30% in some cases.
That range matters. A 10% haircut on a leveraged position during a sudden market move is painful. A 30% loss on assets that then recovered within hours is the kind of thing that drives users away from a protocol, or from DeFi entirely.
To its credit, Aave has acknowledged this tension and taken steps to address it. Aave V4 introduces a redesigned liquidation mechanism that improves on V3’s approach. The adaptive repayment amount means borrowers are not over-liquidated, if only a small amount of debt needs to be repaid to restore a healthy position, that is all the liquidator is allowed to repay.
The variable liquidation bonus also increases as the health factor decreases, creating better incentives for liquidators to act on riskier positions. Whether that meaningfully closes the gap between what’s good for the protocol and what’s good for borrowers remains to be seen in practice.
Recursive leverage and concentrated liquidation waves
Perhaps the most striking finding in the Bank of Canada paper is what was actually driving a large share of borrowing demand on Aave V3 in the first place. Recursive leverage accounted for over 20% of total borrowed volume and 8.2% of borrowing transactions during the sample period.
Recursive leverage involves repeatedly borrowing against collateral, redeploying the borrowed assets as new collateral and borrowing again to amplify exposure. In other words, a meaningful slice of Aave’s usage was amplified speculation, stacked in layers.
That’s not inherently wrong. Leverage is legal, widely used, and embraced by sophisticated participants across both traditional and crypto markets. But it does change the risk profile of the protocol in ways that aren’t always obvious from the outside. When markets move sharply, recursive leverage unwinds fast, and that unwinding tends to compound the very volatility that triggered it.
Liquidations on Aave V3 tended to occur in concentrated waves, with four assets, Wrapped Ether (WETH), Wrapped Staked Ether (wstETH), Wrapped Bitcoin (WBTC), and Wrapped eETH (weETH), accounting for 90% of total liquidated value.
The concentration of that exposure is worth sitting with. It suggests that during a sharp drawdown in any one of those assets, the liquidation cascade could be significant, and the borrowers holding those positions would bear the brunt of it.
This played out in real terms as recently as March 2026. About $27 million worth of borrower positions were liquidated on Aave after what appeared to be a temporary discrepancy in the price of wstETH reported by Aave’s risk oracle, which flagged the token as roughly 2.85% less valuable than it actually was, pushing some borrowing positions below their safety thresholds. The protocol incurred no bad debt. Borrowers weren’t so fortunate.
Since its launch, Aave has processed over 310,000 liquidations totaling $4.65 billion in value, with three-quarters of that volume occurring on Ethereum, where the average liquidation size exceeds $50,000. Those numbers speak to both the scale of the protocol and the frequency with which its liquidation mechanisms are activated.
None of this is to say Aave is broken or poorly designed. The Bank of Canada paper itself concludes that DeFi lending with proper governance is operationally viable, a fairly measured endorsement from a central bank institution that might have been expected to be harsher.
The zero non-performing loan figure in 2024 is genuinely impressive for a fully automated system operating across billions in open positions, without any human credit oversight.
But the paper draws a useful line between protocol health and borrower health. The fact that Aave’s lenders were protected says something positive about the protocol’s design. The fact that borrowers absorbed the losses necessary to make that happen is a separate and equally important question for anyone thinking about DeFi lending’s long-term sustainability and user trust.